- Design
- Retrospective analysis within the NELSON screening trial
- Population
- 213 histologically confirmed screen-detected lung cancers
- Primary outcome
- Association of CT features with adenocarcinoma vs other subtypes
- Effect
- Pleural attachment OR 6.67 (1.20 to 33.33); size OR 0.92 per mm (0.85 to 0.99); AUC 0.75
This retrospective analysis used 213 histologically confirmed lung cancers from the NELSON low-dose CT screening trial. Most participants were men, mean age 64, and two-thirds of the cancers were adenocarcinomas. Nodule size, volume, type, margins, location, volume doubling time and pleural distance were annotated to a standard protocol.
Adenocarcinomas were more often early stage, smaller (14.6 vs 17.3 mm) and in the right upper lobe. In adjusted analysis, larger size was associated with lower odds of adenocarcinoma (OR 0.92 per mm, 95% CI 0.85 to 0.99), pleural attachment with higher odds (OR 6.67, 1.20 to 33.33), and lower and middle lobe location with lower odds. Slower growth was associated with adenocarcinoma. The model's AUC was 0.75, with 96% sensitivity but only 29% specificity.
These associations help explain why a small, slow-growing, pleural-based upper-lobe nodule should not be dismissed on growth alone, because that is the typical adenocarcinoma profile. They do not allow histology to be predicted for an individual nodule; tissue remains necessary.
- Do not reassure on slow growth alone for an upper-lobe or pleural-based nodule; that fits adenocarcinoma.
- Record nodule location, pleural contact and volume doubling time in screening reports.
- Do not infer histology from imaging; the model had only 29% specificity.
- Manage nodules by established screening protocols, not by predicted subtype.
Why it matters
Slow growth is often read as reassurance, but in screening it is a feature of the commonest cancer subtype.
Don't overread it
Imaging features cannot predict histology for an individual patient.
The statistics, in plain English
The odds ratio for pleural attachment, 6.67, has a very wide interval (1.20 to 33.33), meaning the estimate is imprecise. A model with 96% sensitivity and 29% specificity labels most cancers as adenocarcinoma, so it is poor at telling subtypes apart.
Read the rest in the app
You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for radiology, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free